SCAN GUIDE

This study proposes a real-time deep learning system for echocardiographic image quality assessment and probe movement guidance, addressing operator dependency in acquiring standard cardiac views. The system combines multi-task learning models for view classification and segmentation, offering both quantitative quality scores and actionable probe guidance. Quality is assessed using entropy-based uncertainty estimation and structural segmentation metrics, enabling reliable evaluation without extensive manual labeling. The integrated movement prediction model effectively estimates corrective actions based on image features. Experimental results demonstrate high diagnostic performance (AUC: 0.901–0.987; accuracy: 0.820–0.946) and successful correlation between entropy and image quality levels. The model performs especially well in the A4C view (accuracy: 0.845, precision: 0.850), and consistently estimates probe motions like rotation and hold (AP: 0.875–0.998). Overall, the system enhances imaging consistency and reliability, proving its potential for clinical integration and broader application across varied patient populations.